REF-001 · UK · LEGAL
Eliminated 70% of unqualified calls and saved £78k/yr in solicitor time.
A UK employment law practice was losing 10 hours a week to phone triage — 70% of callers had no viable claim. We built a 6-step calculator that scores claims, routes high-value leads to lawyers automatically, classifies inbound email by AI, and handles after-hours enquiries with an AI phone agent. Solicitors now only speak to qualified claimants.
10+ EDGE FUNCTIONS · 3 AI CHANNELS · 30-MIN CALLS → 0
Read the case study → REF-002 · GLOBAL · LOGISTICS
10 production systems across growth, ops, and intelligence.
A global storage marketplace with 100+ locations needed AI across the entire business — not one system, but ten. We built a CRM that sends and sequences outreach from Claude, a location intelligence platform scoring 100 UK cities for expansion, a competitor price tracker scraping 50k datapoints/month, a review management dashboard, a Google Business Profile autoposter, and a link-building pipeline. All in production. All still running.
10 SYSTEMS · 100+ LOCATIONS · 50K DATAPOINTS/MONTH
Read the case study → REF-003 · UK · FINANCIAL SERVICES
1,000× cost reduction on ad creatives. 858 compliant images in two weeks.
A UK financial claims firm was paying £50–150 per ad image from an agency, producing 5–10 per week. We built an AI creative pipeline: Claude writes copy and prompts, Nano Banana generates images, Gemini scores quality, fal.ai produces video, and a two-stage approval workflow catches compliance errors before launch — not after. Cost per image dropped from £50 to £0.05. The first production run delivered 858 compliant creatives across 97 categories.
£0.05/IMAGE vs £50 · 858 IN 2 WEEKS · 2-STAGE COMPLIANCE
Read the case study → REF-004 · UK · ACCOUNTING
£580k/yr in partner time recovered. Tax-season throughput tripled.
A 12-partner UK practice was capping tax-season volume at 240 returns because partners couldn't review faster. We built an automated SA100 + CT600 pipeline: client documents OCR'd and extracted, computations run against UK 2025/26 statutory rates, draft returns presented for partner review in 38 minutes instead of 4 hours. Tax-season throughput hit 850 returns. £580k/yr in partner time recovered.
£180K/YR PARTNER TIME · 850 RETURNS/SEASON · 99.7% ACCURACY
Read the case study → REF-005 · UK · ADVISORY
6-week due diligence in 8 days. 2.5× more deals delivered per year.
Norton Burr handled mid-market M&A mandates on a 6-week DD cycle that was bottlenecked entirely on analyst time spent reading contracts, building schedules, and assembling board packs. We built a Claude-orchestrated DD pipeline: 2,400 deal documents extracted and scored per engagement, financial models auto-populated, comparable deals retrieved from the firm's prior work, board packs generated. Cycle time dropped to 8 days. The firm delivered 18 deals in the next 14 months instead of 8.
6W → 8D DD CYCLE · £840K/YR · 18 DEALS / 14 MONTHS
Read the case study → REF-006 · UK · US · CONSULTING
2,400 engagements indexed. Analyst onboarding 8 weeks → 2 weeks. £1.2M/yr in analyst time recovered.
Olympia Advisory had 2,400 prior engagements sitting in SharePoint, organised by client and date. New analysts spent the first 8 weeks of their tenure learning what work the firm had done. Partners answered the same precedent questions over and over because there was no other way for the team to find the answer. We built a Claude + RAG system over the entire precedent library. Citation-grounded answers in 90 seconds. Analyst onboarding compressed to 2 weeks. £1.2M/yr in analyst time recovered.
2,400 ENGAGEMENTS INDEXED · 87% CITATION HIT RATE · £1.2M/YR
Read the case study → REF-007 · GLOBAL · MARKETPLACE OPS
4 MCP servers replaced 6 SaaS seats + an analyst per quarter.
StowFast's ops + marketing teams needed Uberall, GSC, GA4, and Serpstat data daily. Each platform had a per-seat license; the team funnelled questions through one analyst who built SQL queries on demand. We built 4 read-only MCP servers — each wraps the platform's API, exposes it through Claude.ai's connector model, OAuth-gated. Every team member now queries the data conversationally. No more SQL queues. No more six-seat licenses.
4 MCP SERVERS · 935 LOCATIONS · OAUTH-GATED · CLAUDE.AI CONNECTOR
Read the case study → REF-008 · GLOBAL · PRICING INTELLIGENCE
78 cities. 5 competitors. Pricing decisions in minutes, not weeks.
Northfield Storage operates a global storage marketplace where customer acquisition cost moves with competitor pricing. The ops team needed to know when competitor inventory in a given city was priced 15%+ above or below the floor — and react. We built a tiered scraping pipeline across 5 providers and 78 cities (43 baseline UK/EU + 35 expansion markets), warehoused in Postgres with an MCP server on top, alerts to Slack the morning a city moves outside the band. Pricing decisions that used to take a quarter now happen in minutes.
78 CITIES · 5 PROVIDERS · TIERED SCRAPING · MCP + SLACK ALERTS
Read the case study → REF-009 · UK · ACCOUNTING
12 new regional offices in 18 months. £3.4M attributable revenue. Zero gut-call decisions.
Cardew & Halsey is a UK chartered accountancy network expanding from 14 regional offices to 26 in the medium term. Office choice was historically partner-led — a senior would pick 1-2 candidate cities per year on intuition, validated by a few partner-network conversations. The miss rate was material. We built a market-entry scoring engine: SME density per postcode, FTSE-250 HQ proximity, fee-earner migration patterns, competitor consolidation activity, and HMRC-published practice closure data combined into a ranked shortlist. 12 new offices opened in 18 months, all algorithmically pre-selected. £3.4M in attributable first-year fee revenue.
12 NEW OFFICES · £3.4M REVENUE · 18 MONTHS · ZERO GUT-CALL DECISIONS
Read the case study → REF-010 · UK · PROPERTY · BRAND OPS
220 branches. £150k/yr in SaaS retired. Average rating moved from 3.8 to 4.4.
Greystone Estates runs 220 branches across the UK. Managing reputation at that scale meant either committing to enterprise SaaS at £100-200k/yr, or building it. We built it. A review velocity engine across all 220 branches, brand-mention tracking across PropertyHub, Rightmove community forums, MSE, Twitter, and 30+ property-focused communities, AI compliance filtering on outgoing responses. All self-hosted. £150k/yr in SaaS retired. Average customer rating moved from 3.8 to 4.4 in 11 months.
220 BRANCHES · £150K/YR SAVED · 3.8 → 4.4 AVG RATING
Read the case study → REF-011 · UK · LEGAL · LEGAL-TECH
UK employment-law AI stack at 12× lower LLM cost. Corpus + claim helper + lawyer handoff — production in 11 phases.
Mercer Employment Law needed AI infrastructure across two surfaces: a structured corpus of UK tribunal decisions for retrieval-grounded answers, and a claim-helper agent that triages inbound enquiries and hands off to lawyers when the claim is worth pursuing. The hard constraint was LLM cost — at any meaningful query volume, off-the-shelf Sonnet/Haiku pricing made the unit economics tight. We built the corpus extraction on Gemini Flash-Lite (12× cheaper than Haiku 4.5 for structured-JSON legal extraction at equivalent accuracy), shipped the claim helper across 11 production phases, and wired a frictionless funnel to lawyer review.
12× LLM COST REDUCTION · 11 PHASES SHIPPED · CORPUS + AGENT + FUNNEL
Read the case study →